Gear Tooth Modification for Noise Reduction: A Root-Cause Retrospective
A gearbox that whines is an invoice for rework, and the engineer who opens it often reaches for softer gears or more lubricant before asking the uncomfortable question: where exactly is the excitation coming from? This article is written as a retrospective of a real field case in which a small industrial gearbox unit was returned twice for excessive whine, and how the fix turned out to be neither softer material nor better oil, but a few tenths of a millimetre of metal removed from the right part of the tooth.
Read for the structure more than the numbers. The case shows the four-step discipline that separates gear noise luck from gear noise control: symptom measurement, root-cause separation, modification design, and verification on the rig.
1. Case Background
The unit was a two-stage helical gearbox rated at 7.5 kW, output speed about 140 rpm, driving a small conveyor. The design was conventional: cast housing, induction-hardened pinions, through-hardened wheels, oil splash lubrication. First delivery passed the factory load test within the noise limit of 68 dB(A) at one metre. The customer’s complaint arrived after three weeks of service: a tonal whine at the high-speed stage, loudest at 80% load, disappearing when the unit was unloaded.
Two replacement units were shipped with tighter backlash and better grade tolerances, both measured quieter on the bench and both reported noisy on site. That mismatch between bench and field pointed away from manufacturing scatter and toward excitation that only appeared under real load and real gear deflection.
2. Symptom Measurement Before Any Modification
The team mounted an accelerometer on the bearing housing of the high-speed stage and collected three signals: sound pressure, housing acceleration, and instantaneous shaft speed. The whine frequency resolved to about 2.4 kHz, which matched the meshing frequency of the high-speed pair, 75 teeth times 32 Hz shaft speed. That identification alone answered the first question: the noise was gear mesh excitation, not bearing or housing resonance.
Two more measurements completed the picture. Order analysis showed the mesh tone carried sidebands at the pinion rotational frequency, which points to a once-per-revolution loading variation. And a load sweep from 20% to 100% showed the mesh tone growing faster than linear, the classic signature of tooth deflection changing the contact pattern under load. The culprit was not a manufacturing defect; it was elastic tooth deflection under load creating a hard contact at the tooth end and a correspondingly sharp impact at entry.
3. Root-Cause Separation: Four Possible Excitation Sources
Before touching the gears, the team forced the decision through a four-source checklist, assigning each source a yes-or-no based on the measurements.
| Source | Field Evidence | Verdict |
|---|---|---|
| Transmission error scatter | Bench noise rose with backlash change | Present, minor |
| Tooth flank contacts end | Sidebands at pinion order | Present, dominant |
| Housing resonance | 2.4 kHz far from house modes | Not responsible |
| Lubrication starvation | Stable temperature, no scuffing | Not responsible |
The verdict pointed to the flank: under load the teeth deflected enough that the effective contact pattern ran off the end of the face width at entry and exit, creating an impact each time the pair loaded. That is exactly the failure mode tooth modification is designed to remove.
4. What Tooth Modification Actually Does
Tooth modification is the deliberate, micron-scale removal of material from the flank in two directions. Tip relief takes a small amount off the tooth tip so that when the following tooth engages, the entry contact is softened instead of slamming. Flank or longitudinal crowning makes the contact bulge toward the middle of the face width, so that under bending deflection the contact never runs off the end of the tooth.
| Modification | Direction | Effect On Noise | Typical Size |
|---|---|---|---|
| Tip relief | Radial, at tooth tip | Softens mesh entry impact | 10-40 m |
| Profile crowning | Across profile height | Reduces transmission error | 5-25 m |
| Longitudinal crowning | Along face width | Prevents edge contact | 10-50 m |
| Lead modification | Helix angle adjustment | Compensates shaft deflection | 5-30 m |
The sizes are tiny because the excitation is tiny. Gear noise almost never needs a big design change; it needs a controlled tenth-of-a-millimetre change in exactly the place the deflection concentrates. Get the direction right and the whine drops; get the location wrong and the contact concentrates somewhere even worse.
5. Measurement-Driven Modification Design
In this case the modification was set from measured deflection, not from a textbook average. The team measured the contact pattern using Prussian blue at 20%, 60%, and 100% load. The 100% pattern showed the contact running fully to the end of the face width and a bright line at the tip, both confirming edge and tip impact. The free-body estimate of tooth deflection at full load came to about 18 micrometres at the end of face width.
The chosen modification combined three values. A tip relief of 25 micrometres over the outer 20% of the profile height, a longitudinal crowning of 20 micrometres over the full face width, and a lead change of 12 micrometres on the pinion to offset housing bore deflection. All three were specified on the corrected gear drawing and machined on the same hobbing and grinding equipment that produced the originals.
6. Implementation and the Number That Mattered
The corrected gears were cut and ground, then re-run through the identical acceptance sequence that the original units failed on site. This time the results were decisive. The whine at the identified frequency dropped from 68 dB(A) to 61 dB(A) at full load, a 7 dB reduction that corresponds to roughly halving the perceived loudness besides the gear mesh. The floor, not the bench, was the judge: the same customer conveyor reported the unit within the site limit under identical load.
Three qualitative confirmations accompanied the number. The mesh tone no longer grew faster than linear across the load sweep, indicating the impact at entry had been removed. The sidebands at pinion order almost vanished, showing the once-per-revolution loading variation was addressed by the crowning. And a Prussian-blue check at 100% load showed the contact pattern safely inside the face width, with a clean elliptical patch centred in the tooth.
Retrospective lesson: the two replacement units shipped before the root-cause analysis were cheaper to make than the analysis, and more expensive to the company than both. The measurement time paid for itself in the first corrected batch.
7. The Reusable Four-Step Noise Discipline
Any gear noise complaint can be run through the same sequence that solved this case.
- Identify the source frequency by order analysis; a mesh-frequency match proves the gears excite it.
- Separate the cause with the source checklist, assigning evidence, never assumptions.
- Measure the load-dependent contact pattern and deflection before specifying modification values.
- Verify on the load sweep and the customer installation, not only on the bench.
Follow the discipline and tooth modification becomes a controlled design action. Skip any step and it becomes another expensive attempt at noise reduction by guesswork.
8. When Modification Is Not the Answer
Tooth modification is powerful but not universal, and knowing when to stop prevents wasted grinding time. Modification will not fix noise from a structurally resonant housing, a misaligned bearing seat, or a gear with generational wear. It also reaches diminishing returns: beyond a few tens of micrometres of relief, additional metal removal reduces load capacity faster than it reduces noise.
| Diagnosis | Modification? | Correct Direction |
|---|---|---|
| Edge contact under load | Yes | Crown the flank |
| Tip impact at entry | Yes | Tip relief |
| Housing resonance at mesh order | No | Rib, damp, or change stiffness |
| Material or heat-treatment distortion | Limited | Fix process control |
| Bearing housing misalignment | No | Correct bore or fits |
The table keeps the discipline honest. Modification is a precision adjustment on a system that is otherwise sound, not a repair that compensates for loose tolerances elsewhere in the drive train.
9. Conclusion and the Cost of the Lesson
The retrospective ends where it began: the returned gearbox was not a material problem or an oil problem, it was a geometry problem that only appeared under load. Three batches of corrected units have now run in the field without a single whine complaint, and the same analysis folder has been reused twice on larger units with the same methodology and similar success.
The takeaway for any gear designer is compact. Measure the excitation before changing the gear. Separate the sources on evidence. Size the modification from the measured deflection and contact pattern. And verify where the customer listens, not just where the test rig is comfortable. Tooth modification is a surgical tool, and like any surgical tool, its value is decided by the diagnosis that precedes it.
10. The Measurement Toolkit Behind the Numbers
The modification values in this retrospective were not pulled from a chart; they came from four instruments that every gear shop can access. A Prussian-blue contact print under load showed where the flank really touched. A single-flank roll tester gave the composite transmission error, isolating the mesh sinuosity that drives whine. Order analysis software on a vibration analyser separated the mesh tone from its sidebands. And an encoder on the input shaft measured instantaneous angular velocity ripple, exposing the once-per-revolution loading variation that flank crowning would cancel.
Use the instruments in this order: contact print to see the pattern, roll test to quantify the error, order analysis to identify the dominant order, and speed ripple to trace the source event. Each instrument narrows the diagnosis by exactly one step, and together they leave almost no room for the expensive guess.
11. Reading a Gear Drawing After Modification
Once the modification is specified, the drawing must carry it unambiguously or the shop will grind it back to a plain involute. Check for four lines of information. The profile form must show the tip relief amount and the length of the relief zone as a fraction of the profile height. The lead or crowning value must be stated as total drop over the face width, not as a radius that confuses the inspector. The measurement datum must identify which side of the tooth the values are referenced to. And the acceptance tolerance must state the allowed deviation, commonly plus or minus 3-5 micrometres on the relief amount.
A correctly annotated drawing converts a hard-won diagnosis into reproducible production. An ambiguous one re-opens the old discussion next season, when a different operator reads the same symbol differently and the whine quietly returns.
12. Glossary
| Term | Meaning in Gear Engineering |
|---|---|
| Tip relief | Micron-scale removal at the tooth tip to soften entry |
| Crowning | Longitudinal bulge that prevents edge contact |
| Transmission error | Deviation between theoretical and actual gear output rotation |
| Order analysis | Vibration spectrum resolved into shaft-order harmonics |
| Mesh frequency | Teeth count times shaft speed, the base whine harmonic |
| Contact pattern | Visible flank touch area under applied load |
Armed with the case, the measurement sequence, and the drawing requirements from this article, the gear noise problem moves from a customer complaint to a design calculation. That is the entire point of the retrospective: every decibel of whine has a diagnosable root, and a surgical modification, applied after that diagnosis, is how you remove it.
13. A Field Acceptance Checklist for Quiet Gears
Close the retrospective with the checklist that the corrected batch used before release.
- Confirm the order analysis shows the mesh tone, not a housing mode, as the dominant source.
- Verify the contact pattern stays inside the face width from no-load to full load.
- Confirm tip relief and crowning values match the drawing tolerance, plus or minus a few micrometres.
- Run a load sweep and record the slope of the mesh tone; it must stay near linear.
- Log the Prussian-blue pattern at full load into the quality file for the next build.
- Document the customer-side reading at rated load before sign-off.
Check the boxes in order and the next gearbox leaves the floor with the same confidence as the corrected batch in this case, not quieter by luck, but quieter by design.
